Scott E. Nielsen

Evaluating resource selection functions

A resource selection function (RSF) is any model that yields values proportional to the probability of use of a resource unit. RSF models often are fitted using generalized linear models (GLMs) although a variety of statistical models might be used. Information criteria such as the Akaike Information Criteria (AIC) or Bayesian Information Criteria (BIC) are tools that can be useful for selecting a model from a set of biologically plausible candidates. Statistical inference procedures, such as the likelihood-ratio test, can be used to assess whether models deviate from random null models. But for most applications of RSF models, usefulness is evaluated by how well the model predicts the location of organisms on a landscape. Predictions from RSF models constructed using presence/absence (used/ unused) data can be evaluated using procedures developed for logistic regression, such as confusion matrices, Kappa statistics, and Receiver Operating Characteristic (ROC) curves. However, RSF models estimated from presence/ available data create unique problems for evaluating model predictions. For presence/available models we propose a form of k -fold cross validation for evaluating prediction success. This involves calculating the correlation between RSF ranks and area-adjusted frequencies for a withheld sub-sample of data. A similar approach can be applied to evaluate predictive success for out-of-sample data. Not all RSF models are robust for application in different times or different places due to ecological and behavioral variation of the target organisms. # 2002 Elsevier Science B.V. All rights reserved.

Resource Selection Functions Based on Use–Availability Data: Theoretical Motivation and Evaluation Methods

Abstract Applications of logistic regression in a used–unused design in wildlife habitat studies often suffer from asymmetry of errors: used resource units (landscape locations) are known with certainty, whereas unused resource units might be observed to be used with greater sampling intensity. More appropriate might be to use logistic regression to estimate a resource selection function (RSF) tied to a use–availability design based on independent samples drawn from used and available resource units. We review the theoretical motivation for RSFs and show that sample “contamination” and the exponential form commonly assumed for the RSF are not concerns, contrary to recent statements by Keating and Cherry (2004; Use and interpretation of logistic regression in habitat-selection studies. Journal of Wildlife Management 68:774–789). To do this, we re-derive the use–availability likelihood and show that it can be maximized by logistic regression software. We then consider 2 case studies that illustrate our findings. For our first case study, we fit both RSFs and resource selection probability functions (RSPF) to point count data for 4 bird species with varying levels of occurrence among sample blocks. Drawing on our new derivation of the likelihood, we sample available resource units with replacement and assume overlapping distributions of used and available resource units. Irrespective of overlap, we observed approximate proportionality between predictions of a RSF and RSPF. For our second case study, we evaluate the classic use-availability design suggested by Manly et al. (2002), where availability is sampled without replacement, and we systematically introduce contamination to a sample of available units applied to RSFs for woodland caribou (Rangifer tarandus caribou). Although contamination appeared to reduce the magnitude of one RSF beta coefficient, change in magnitude exceeded sampling variation only when >20% of the available units were confirmed caribou use locations (i.e., contaminated). These empirically based simulations suggest that previously recommended sampling designs are robust to contamination. We conclude with a new validation method for evaluating predictive performance of a RSF and for assessing if the model deviates from being proportional to the probability of use of a resource unit.

SEASONAL AND DIEL PATTERNS OF GRIZZLY BEAR DIET AND ACTIVITY IN WEST-CENTRAL ALBERTA

Abstract Seasonal food habits and activity patterns were examined for grizzly bears (Ursus arctos) in west-central Alberta, Canada, to better understand habitat requirements in a threatened population. Food habits were based on an analysis of 665 feces collected from 18 grizzly bears between April and October 2001–2003. Trends in the use of foods were comparable to those of other central Rocky Mountain populations, with minor differences likely reflecting regional habitat and forage availability. Five activities (bedding, sweet vetch digging, insect feeding, frugivory, and ungulate kills) were identified for each of 1,032 field-visited global positioning system radiotelemetry locations from 9 female grizzly bears. We predicted the probability of each activity during relevant periods by time of day (crepuscular, diurnal, and nocturnal) and habitat. Selection ratios were used to assess which habitat and time periods were selected. Activity patterns changed considerably over a 24-h period, with foraging activities occurring mostly during diurnal and crepuscular periods and bedding at night. Habitats were important predictors of activity. Forested areas were selected for bedding areas, whereas digging, insect-foraging, and frugivory activities were associated with herbaceous, recently disturbed forest and open-canopy forests. We suggest that researchers consider behavior and time of day in analyses of habitat selection to improve explanations of habitat use and mechanisms of selection.

Development and testing of phenologically driven grizzly bear habitat models

Abstract We developed and compared three habitat models for estimating the relative probability of occurrence, by month, for grizzly bears (Ursus arctos) in Jasper National Park (JNP), Alberta. These models included 1) a habitat map derived from remote sensing Landsat imagery; 2) food-index models generated from the predicted occurrence of bear foods and assigned monthly importance values; and 3) probabilistic food models representing the occurrence of each bear food. Resource selection function (RSF) models for grizzly bears were generated using 3,924 global positioning system (GPS) radiotelemetry locations and the above habitat models. Comparisons were made among RSF models, by month, using Akaike’s Information Criterion (AIC). In all seven months (April to October), food-index models performed poorly. In April and July, the remote-sensing habitat map predicted bears best, while the food-probability models performed best in the remaining five months. Overall, we found substantial improvement by using food-probability models for predicting JNP grizzly bear occurrence. Remote-sensing maps, although predictive, may not reveal underlying mechanisms and fail to recognize the dynamic nature of seasonal grizzly bear habitats. The disconnect between food-index and food-probability models suggests that monthly food importance values require additional parameterization. Development of spatial food models on phenologically important scales more closely matches the resources and temporal scales at which animals perceive and use resources.

Velocity of climate change algorithms for guiding conservation and management.

The velocity of climate change is an elegant analytical concept that can be used to evaluate the exposure of organisms to climate change. In essence, one divides the rate of climate change by the rate of spatial climate variability to obtain a speed at which species must migrate over the surface of the earth to maintain constant climate conditions. However, to apply the algorithm for conservation and management purposes, additional information is needed to improve realism at local scales. For example, destination information is needed to ensure that vectors describing speed and direction of required migration do not point toward a climatic cul-de-sac by pointing beyond mountain tops. Here, we present an analytical approach that conforms to standard velocity algorithms if climate equivalents are nearby. Otherwise, the algorithm extends the search for climate refugia, which can be expanded to search for multivariate climate matches. With source and destination information available, forward and backward velocities can be calculated allowing useful inferences about conservation of species (present-to-future velocities) and management of species populations (future-to-present velocities).

Linking ground-based to satellite-derived phenological metrics in support of habitat assessment

Changes in the timing of plant phenology are important indicators of inter-annual climatic variations and are a critical driver of food availability and habitat use for a range of species. A number of remote sensing techniques have recently been developed to observe vegetation cycles throughout the year, including the use of inexpensive visible spectrum digital cameras at the stand level and the use of high temporal frequency Advanced Very High Resolution Radiometer National Oceanic and Atmospheric Administration (AVHRR NOAA) and MODerate resolution Imaging Spectroradiometer (MODIS) imagery at a satellite scale. A fundamental challenge with using satellite data to track plant phenology, however, is the trade-off between the level of spatial detail and the revisit time provided by the sensor, and the ability to verify the interpretation of phenological activity. One way to address this challenge is to integrate remotely sensed observations obtained at different spatial and temporal scales to provide information that contains both high temporal density and fine spatial resolution observations. In this article, we compare measures of vegetation phenology observed from a network of ground-based cameras with satellite-derived measures of greenness derived from a fused broad (MODIS) and fine spatial (Landsat) scale satellite data set. We derive and compare three key indicators of phenological activity including the start date of green-up, start date of senescence and length of growing season from both a ground-based camera network and 30 m spatial resolution synthetic Landsat scenes. Results indicate that although field-based estimates, generally, predicted an earlier start and end of the vegetation season than the fused satellite observations, highly significant relationships were found for the prediction of the start (R 2 = 0.65), end (R 2 = 0.72) and length (R 2 = 0.70) of the growing season across all sites. We conclude that some predictable bias exists however unlike visual field measures of the collected data represent both a spectral and a visual archive for later use.

Macronutrient optimization and seasonal diet mixing in a large omnivore, the grizzly bear: a geometric analysis.

Nutrient balance is a strong determinant of animal fitness and demography. It is therefore important to understand how the compositions of available foods relate to required balance of nutrients and habitat suitability for animals in the wild. These relationships are, however, complex, particularly for omnivores that often need to compose balanced diets by combining their intake from diverse nutritionally complementary foods. Here we apply geometric models to understand how the nutritional compositions of foods available to an omnivorous member of the order Carnivora, the grizzly bear (Ursus arctos L.), relate to optimal macronutrient intake, and assess the seasonal nutritional constraints on the study population in west-central Alberta, Canada. The models examined the proportion of macronutrients that bears could consume by mixing their diet from food available in each season, and assessed the extent to which bears could consume the ratio of protein to non-protein energy previously demonstrated using captive bears to optimize mass gain. We found that non-selective feeding on ungulate carcasses provided a non-optimal macronutrient balance with surplus protein relative to fat and carbohydrate, reflecting adaptation to an omnivorous lifestyle, and that optimization through feeding selectively on different tissues of ungulate carcasses is unlikely. Bears were, however, able to dilute protein intake to an optimal ratio by mixing their otherwise high-protein diet with carbohydrate-rich fruit. Some individual food items were close to optimally balanced in protein to non-protein energy (e.g. Hedysarum alpinum roots), which may help explain their dietary prevalence. Ants may be consumed particularly as a source of lipids. Overall, our analysis showed that most food available to bears in the study area were high in protein relative to lipid or carbohydrate, suggesting the lack of non-protein energy limits the fitness (e.g. body size and reproduction) and population density of grizzly bears in this ecosystem.

Does Learning or Instinct Shape Habitat Selection

Habitat selection is an important behavioural process widely studied for its population-level effects. Models of habitat selection are, however, often fit without a mechanistic consideration. Here, we investigated whether patterns in habitat selection result from instinct or learning for a population of grizzly bears (Ursus arctos) in Alberta, Canada. We found that habitat selection and relatedness were positively correlated in female bears during the fall season, with a trend in the spring, but not during any season for males. This suggests that habitat selection is a learned behaviour because males do not participate in parental care: a genetically predetermined behaviour (instinct) would have resulted in habitat selection and relatedness correlations for both sexes. Geographic distance and home range overlap among animals did not alter correlations indicating that dispersal and spatial autocorrelation had little effect on the observed trends. These results suggest that habitat selection in grizzly bears are partly learned from their mothers, which could have implications for the translocation of wildlife to novel environments.

Restoration of Midwest Oak Barrens: Structural Manipulation or Process-only?

We investigated vegetation responses in terms of canopy, ground-layer diversity, and ecological species groups using two restoration treatments at two degraded oak barren and savanna sites in central Wisconsin, USA. The two restoration models tested were (1) process-only, which reintroduced fire in the form of prescribed burning, and (2) structural manipulation, which used prescribed burning following selective timber removal. Both methods have been widely promoted, debated, and investigated in the fire-prone ecosystems of western North America, but they have not been studied in midwestern ecosystems. Vegetation was monitored in permanent quadrats prior to and following treatment applications. All treatment responses were compared against trends at control sites. We used diversity, canopy, and cover estimates within ecological groups between pre- and post-treatment periods as our response. Effect size was calculated, and the statistical significance of effects was determined using one-factor analysis of variance. Following treatments, canopy levels were restored to prior savanna levels with structural manipulation, but failed to respond to process-only approaches. Likewise, multiple positive responses were detected in the ground layer with structural manipulation, but few with process-only treatments. Despite initial responses, ground-layer restoration appears to be constrained by the dominance of Pennsylvania sedge (Carex pensylvanica). Many savanna forbs, legumes, and C4 graminoids were missing. We presume that 70 yr of fire suppression and associated succession to oak woodlands were largely responsible for sedge conversion and the loss of savanna species. Despite observed limitations, structural manipulation treatments appeared to be more effective than process-only approaches. Sites with holdover savanna species that have not been dominated by sedge should be targeted for immediate restoration before further losses occur. Further investigation of sedge mat thresholds and long-term restoration dynamics is required.

Spatial and Temporal Heterogeneity Creates a “Brown Tide” in Root Phenology and Nutrition

Spatial and temporal heterogeneity in plant phenology and nutrition benefits herbivores by prolonging the period in which they can forage on nutritious plants. Landscape heterogeneity can therefore enhance population performance of herbivores and may be a critically important feature of their habitat. The benefits of resource heterogeneity over space and time should extend not only to large herbivores using above-ground vegetation but also to omnivores that utilize below-ground resources. We used generalized linear models to evaluate whether spatial heterogeneity influenced temporal variation in the crude protein content of alpine sweetvetch (Hedysarum alpinum) roots in west-central Alberta, Canada, thereby potentially offering nutritional benefits to grizzly bears (Ursus arctos). We demonstrated that temporal patterns in the crude protein content of alpine sweetvetch roots were influenced by spatial heterogeneity in annual growing season temperatures and soil moisture and nutrients. Spatial heterogeneity and asynchrony in the protein content of alpine sweetvetch roots likely benefit grizzly bears by prolonging the period they can forage on high quality resources. Therefore, we have presented evidence of what we termed a “brown wave” or “brown tide” in the phenology and nutrition of a below-ground plant resource, which is analogous to the previously described “green wave” in above-ground resources.